Boy, that last post was kind of a downer, wasn’t it? Before I drive all the traffic away with depressing posts, maybe I can get back to hifi and get people thinking again.
Glad I stirred things up with the filament-power post - it even chased an old friend out of the woodwork. (Hey Phil! Give me a shout! I live in Colorado now!)
What I didn’t mention were some of subtler issues of filament power supplies - most importantly, the difference between differential and common-mode noise rejection, which are treated very differently by a filament-powered vacuum tube.
Let’s review how a direct-heated triode works: it amplifies the voltage difference between the grid and virtual cathode. The triode does NOT care about ground or the plate voltage, or any other part of the circuit. The virtual cathode is the electrical center of the filament, which is pretty well balanced - it has 50dB or more rejection of the 2.5, 4, or 5VAC that powers the filament.
This has implications for the filament power supply - noise that rides in differential form is strongly rejected, while anything that shakes BOTH wires has NO rejection at all. Think about it: anything that moves both sides of the filament is amplified just as much as a grid signal!
Even for a simple AC supply, there are implications: noise from the raw AC line will ride right in - moving both sides of the filament - unless special steps are taken to electrostatically screen the filament transformer. If the filament is powered with a winding from the main power transformer, stray capacitance will couple rectification noise pulses from the B+ circuit into the much more sensitive low-voltage circuit. The worst-case scenario employs generic solid-state rectifiers for the B+ rectification and NO electrostatic screening for the AC filament supply - unfortunately, this is exactly how many commercial high-end triode amplifier are built!
The quick-n-dirty “solution” to the inevitable hash and high noise floor of the previous circuit is a DC supply for the filament, typically done with a 10V or so AC supply, a solid-state rectifier bridge, a 1000uF “smoothing” cap, and a 3-pin integrated-circuit voltage regulator. Unfortunately, although this measures fairly well, the measurement measures the wrong thing: it doesn’t capture the noise that appears on BOTH sides of the filament, which the noise that matters to the triode.
All the 3-pin regulator does is reduce differential noise, which is already strongly rejected by the triode. It has NO effect on the common-mode noise, the one that matters, so any noise from the rectifier circuit - or AC line noise that is capacitively coupled through the transformer - sails right into the triode, despite the superficially good measurements. Remember, the entire filament supply is floating up around 50 to 80 volts, and the loop area of the supply acts as an antenna for noise pickup from rectifiers elsewhere in the amplifier. A badly laid-out amplifier could have a loop area of several feet, if the DC supply is on one side and the triodes are on the other side. (The loop area is everything between the triode socket and the AC-powerline side of the supply.)
A better, and more comprehensive solution are chokes in series with the filament - these are desirable for both AC and DC supplies, to improve rejection of high-frequency noise that appears on both sides of the filament. Yes, this means two chokes, or a special common-mode choke, and of course the DC resistance must be a fraction of an ohm, otherwise the filament power is wasted heating the choke. Fortunately, impedances are so low the choke can be in the millihenry range (think subwoofer inductor), but it’s still not going to be compact or light.
To be continued …

I’m currently building a SET amp with an all solid state power supply. I have studied the amplifier designs on your website and I realise that it will take a lot of effort to minimize the rectifier noise. To get a better understanding of the subject and to optimize the PSU design I used the DuncanAmpsPSU Designer. The simulations for the heater supply can be found Why solid state supply? To see if it can be done.
Please excuse my ignorance, somehow I managed to mess up the text and link.
1,000 uf filter cap sounds rather small to me.
Most 300B amp designs I have seen use a simple CAP
filter like 10,000 uf or bigger.
But all this is minor talk here, I am re-building a
6B4G SET amp that is 6.3 V @ 1.0 amp, and I need
a good filter design about 2 weeks ago, I just ordered
all my parts back then. With my old design I used
a non-regulated DC power supply with 9,400 uf filter
cap as a hack, as the orginal design used AC heating and
a two 47 ohm resistors to simulate a 100 ohm pot.
This gives me about .5 V ripple.
This my new design for a 1.a amp load,A simple RC
filter: 8.5 V DC ( 2 volt ripple )
to a 3,300 uf cap , 2 1 ohm resistors , 4700 uf to
the filament of the 6B4G. Duncan Amps PSU I get about
.15 volts ripple.
I would have liked to have used a choke
but size of my chassis is only 17″ x 11″ and I don’t
have the room. The RC filter has one advantage is that
it provides about a 2 volt drop giving me some current limiting
to the GB4G’s while the filments are cold.
I may add a two 3.3 volt 5 watt zener diodes to clamp
the voltage out to 6.6 volts, since I don’t like the idea
of using 3 terminal regulator because of negitive feedback
used in the error amp on filaments.
This is a Mid-Fi project since I am not using a 300B
tube and for space reasons Si diodes/Zener diodes/
Leds everywhere for power supplies and bias.I hope
I got the snubber/bypass caps right - 2.5 uf/ 120 vac film
and they fit just after the 7.7V 1.1 amp transformer
and the 4700uf cap since I am trying to make it all fit
but have as many improvements as possible.
What takes all the room in the chassis for this stereo amp
is the regulated valve power supply. The only feedback loop
is in the power supply and that I don’t think is problem.
Power supplies is one item neglected on your basic
2A3 amp. I feel ripple in simple power supplies gets more
attention rather than a real problem of good load and line
regulation. Once you get a clean supply then you can start
cleaning up your design for the best sound.
I had more ro less the same idea, start with a stable, clean power supply and take it from there. In my design I use HEXFREDs, a classic cap-choke-cap filter and a high voltage regulator. The regulator is a design by John Broskie, it is a floating (series)regulator which has its own power supply.
The hum & noise on the high voltage supply is currently -72dBV or 250µV under full load (500V, 100mA). If you power the regulator with two 9V batteries the noise drops to -99dBV or 11µV.
The details and measurements (noise spectra) can be found on my website:
Hands up those who have tried lead-acid batteries for DC filament supply.